If you’ve never done research before, you probably imagine standout conclusions and polished results. That wasn’t my experience.
There Was Supposed to Be an Answer. permalink
I first began working as a research assistant roughly one year ago. The work I pursued had a singular focus: streamline the surface defect detection process using AI technology. More specifically, the goal was to program deep learning models capable of accurately detecting the hidden flaws inside metal (such as cracks or pores). The work aimed to lay a foundation that scientists could later build on through further research.
With no prior experience on the subject, I was unaware of what awaited me. Like most students, my understanding of problem solving was historically shaped by my coursework. Most assignments had a defined solution, straightforward and linear objectives, and a clear manner of gauging progress. So research would work the same way too, right?
Instead of asking, “What is the correct answer?” I found myself instead having to ask, “What can I learn from the information I currently have?” My goal slowly transitioned from trying to arrive at a definite conclusion to now investigating a question whose outcome remained uncertain.
Only now was my progress not measured by whether I reached the answer. Progress was measured by whether I understood the problem a little better than I did before.
If we knew what it was we were doing, it would not be called research, would it? — Albert Einstein
Getting Up to Speed. permalink
For the first month or so, my research mainly consisted of building the foundation I needed for later work. Literature reviews quickly became a staple of my routine. Over time, each paper I read seemed to merge endlessly into the next. It was during this time that I felt especially behind. I would read a paper and need to look up the terms unfamiliar to me, only to find that each explanation introduced even more jargon.
At the same time, I was teaching myself MATLAB and Python scripting (to learn how deep learning models like ResNet and GAN actually worked). Most often than not, this typically resulted in countless late nights.
Spending this time learning didn’t feel especially productive. After weeks of reading papers and teaching myself to code, I still felt overwhelmed and uncertain about whether I was making progress at all. I kept waiting for the moment when I would finally feel caught up. That moment never really came.
I routinely struggled to understand technical jargon and often moved at a snail’s pace through papers. That hasn’t really changed even now. However, despite not knowing everything, I became more comfortable using the knowledge I did have to guide my research.
The Finish Line Kept Moving. permalink
Many times in the research process, what initially felt like the conclusion became the starting point for deeper examination. Once our deep learning model achieved the results we were targeting, the focus then shifted toward understanding why it performed the way it did and whether it could be improved further.
As our research progressed, changes to our model became increasingly rare. However, the challenge wasn't necessarily recognizing that progress had slowed. Determining whether meaningful improvements were possible remained a process not straightforward and continually vague.
Eventually, we stopped once we had exhausted the options likely to meaningfully improve the model’s performance, and the remaining ones were no longer worth the time (or effort frankly) required to explore them.
Drawing the Line. permalink
My work ultimately led to the publication of a conference paper in MSEC (Manufacturing Science and Engineering Conference 2026) and a journal article in Machines. But more than the outcomes, the experience changed how I understand the process of research itself.
Research rarely ends in a clean or definitive way. Instead, it reaches a point where progress becomes painstakingly slow and decisions about whether to continue are based on diminishing returns or practical constraints (such as time or funding).
And over time, I realized research is really just about figuring out what to do next when the answer isn’t fully there yet.